def slow_thin(binimg):
"""
This was the old implementation
"""
from mahotas.bbox import bbox
from mahotas._morph import hitmiss
_struct_elems = []
_struct_elems.append([
[0,0,0],
[2,1,2],
[1,1,1]])
_struct_elems.append([
[2,0,0],
[1,1,0],
[1,1,2]])
_struct_elems.append([
[1,2,0],
[1,1,0],
[1,2,0]])
_struct_elems.append([
[1,1,2],
[1,1,0],
[2,0,0]])
_struct_elems.append([
[1,1,1],
[2,1,2],
[0,0,0]])
_struct_elems.append([
[2,1,1],
[0,1,1],
[0,0,2]])
_struct_elems.append([
[0,2,1],
[0,1,1],
[0,2,1]])
_struct_elems.append([
[0,0,2],
[0,1,1],
[2,1,1]])
_struct_elems = [np.array(elem, np.uint8) for elem in _struct_elems]
res = np.zeros_like(binimg)
min0,max0,min1,max1 = bbox(binimg)
r,c = (max0-min0,max1-min1)
image_exp = np.zeros((r+2, c+2), np.uint8)
imagebuf = np.zeros((r+2,c+2), np.uint8)
prev = np.zeros((r+2,c+2), np.uint8)
image_exp[1:r+1, 1:c+1] = binimg[min0:max0,min1:max1]
while True:
prev[:] = image_exp[:]
for elem in _struct_elems:
newimg = hitmiss(image_exp, elem, imagebuf)
image_exp -= newimg
if np.all(prev == image_exp):
break
res[min0:max0,min1:max1] = image_exp[1:r+1, 1:c+1]
return res
def _compute_features(img):
bimg = (img > 0)
s00,s01,s10,s11 = bbox(bimg)
if s00 > 0: s00 -= 1
if s10 > 0: s10 -= 1
bimg = bimg[s00:s01+1,s10:s11+1]
hull = convexhull(bimg - pymorph.erode(bimg))
Allfeats = np.r_[features.hullfeatures.hullfeatures(img,hull),features.hullfeatures.hullsizefeatures(img,hull)]
return Allfeats[np.array([0,1,2,3,5,6,7],int)]
def dist_hausdorff_compare(segmented, ref):
joint = np.histogram2d(ref.ravel(), segmented.ravel(),
np.arange(max(ref.max(), segmented.max()) + 2))[0]
assignments = joint.argmax(0)[:segmented.max() + 1]
#revassignments = joint.argmax(1)[:ref.max()+1]
values = []
hausdorff = []
for i in xrange(1, int(segmented.max() + 1)):
if assignments[i] == 0:
values += ['bg']
continue
min1, max1, min2, max2 = bbox((ref == assignments[i])
| (segmented == i))
segc = (segmented[min1:max1, min2:max2] == i)
refc = (ref[min1:max1, min2:max2] == assignments[i])
dref = np.maximum(ndimage.distance_transform_edt(refc),
ndimage.distance_transform_edt(~refc))
values.append(dref[segc != refc].sum() /
float(dref[segc | refc].sum()))
hausdorff.append((dref * (segc != refc)).max())
return values, hausdorff
def slow_thin(binimg):
"""
This was the old implementation
"""
from mahotas.bbox import bbox
from mahotas._morph import hitmiss
_struct_elems = []
_struct_elems.append([[0, 0, 0], [2, 1, 2], [1, 1, 1]])
_struct_elems.append([[2, 0, 0], [1, 1, 0], [1, 1, 2]])
_struct_elems.append([[1, 2, 0], [1, 1, 0], [1, 2, 0]])
_struct_elems.append([[1, 1, 2], [1, 1, 0], [2, 0, 0]])
_struct_elems.append([[1, 1, 1], [2, 1, 2], [0, 0, 0]])
_struct_elems.append([[2, 1, 1], [0, 1, 1], [0, 0, 2]])
_struct_elems.append([[0, 2, 1], [0, 1, 1], [0, 2, 1]])
_struct_elems.append([[0, 0, 2], [0, 1, 1], [2, 1, 1]])
_struct_elems = [np.array(elem, np.uint8) for elem in _struct_elems]
res = np.zeros_like(binimg)
min0, max0, min1, max1 = bbox(binimg)
r, c = (max0 - min0, max1 - min1)
image_exp = np.zeros((r + 2, c + 2), np.uint8)
imagebuf = np.zeros((r + 2, c + 2), np.uint8)
prev = np.zeros((r + 2, c + 2), np.uint8)
image_exp[1:r + 1, 1:c + 1] = binimg[min0:max0, min1:max1]
while True:
prev[:] = image_exp[:]
for elem in _struct_elems:
newimg = hitmiss(image_exp, elem, imagebuf)
image_exp -= newimg
if np.all(prev == image_exp):
break
res[min0:max0, min1:max1] = image_exp[1:r + 1, 1:c + 1]
return res
def preprocessimage(image, regionid=None, crop=True, options = {}):
"""
Preprocess the image
image should be an Image object
regionid should be an integer, which indexes into image.region
options is a dictionary. The following options are accepted:
+ 'bgsub.way': One of
- 'ml' (default)
- 'mb' and controls whether the region is selected prior to preprocessing
+ '3d.mode': One of
- 'perslice' (default): process each slice separately
+ 'threshold.algorithm': One of
- 'rc': Riddlar-Calvard
- 'mean': Image mean
"""
def preprocessimg(img):
if len(img.shape) > 2:
assert len(img.shape) == 3, "Cannot handle images of more than 3 dimensions."
if options.get('3d.mode','perslice') == 'perslice':
nr_slices=img.shape[0]
out_proc=img.copy()
out_res=img.copy()
for z in xrange(nr_slices):
proc,res=preprocessimg(img[z])
out_proc[z]=proc
out_res[z]=res
return out_proc,out_res
else:
raise Exception('pyslic.preprocessimg: Do not know how to handle 3d.mode: %s' % options['3d.mode'])
if do_bgsub:
regions = image.regions
img = img.copy()
if regions is not None:
rid = (regionid if regionid is not None else 1)
if options.get('bgsub.way','ml') == 'ml':
img *= (regions == rid)
img = bgsub(img, options)
else:
img = bgsub(img, options)
img *= (cropimg == rid)
else:
if regionid:
warn('Selecting a region different from 1 for an image without region information')
img = bgsub(img, options)
imgscaled = stretch(img, 255)
T = thresholdfor(imgscaled,options)
mask = (imgscaled > T)
mask = majority_filter(mask)
residual = img.copy()
img *= mask
residual *= ~mask
return img,residual
image.lazy_load()
protein = image.channeldata['protein']
dna = image.channeldata.get('dna')
do_bgsub = True
if 'bgsubprotein' in image.temp:
protein = image.temp['bgsubprotein']
do_bgsub = False
if regionid is not None and 'region_ids' in image.temp:
location = image.temp['region_ids'][regionid - 1]
if location is None:
image.channeldata['procprotein'] = \
image.channeldata['resprotein'] = \
image.channeldata['procdna'] = np.zeros((0,0), dtype=protein.dtype)
return
protein = protein[location]
if dna is not None:
dna = dna[location]
image.channeldata['procprotein'],image.channeldata['resprotein'] = preprocessimg(protein)
if dna is not None:
image.channeldata['procdna'],_ = preprocessimg(dna)
if crop:
fullimage = (image.channeldata['procprotein'] > 0) | (image.channeldata['resprotein'] >0)
if 'dna' in image.channeldata:
fullimage |= (image.channeldata['procdna'] > 0)
min1,max1,min2,max2 = bbox(fullimage)
border = 2
min1 = max(0, min1 - border)
min2 = max(0, min2 - border)
max1 += border
max2 += border
image.channeldata['procprotein'] = image.channeldata['procprotein'][min1:max1,min2:max2]
image.channeldata['resprotein'] = image.channeldata['resprotein'][min1:max1,min2:max2]
if 'dna' in image.channeldata:
image.channeldata['procdna'] = image.channeldata['procdna'][min1:max1,min2:max2]
